Introduces the -name-bind frontend action that is intended as an intermediary between the parse-only actions and a full typechecking pass. In this phase, module imports will be validated and resolved, making it possible to emit full make-style dependencies files among other things.
Note that all information available to a parse-only pass is available to name binding, but because it does not continue-on to typecheck input files, full semantic information is not.
TBD validation is effectively an expensive assertion, and is currently only
tuned for Apple platforms. However, we don't want it to regress more, and it
would be nice to start getting validation from people using master
snapshots. Together, this means that turning it on by default for the cases
mentioned above is an appropriate course of action.
At the very least, this has the benefit of running validation across the stdlib,
the overlays and the whole testsuite on each build, so people making changes to
the compiler that change symbols are hopefully alerted.
One limitation here is that this is only validating that the TBD is a superset
of the true set of symbols: it could include spurious symbols that aren't
actually in the binary. This case is less problematic for Swift than symbols
missing from the TBD file, and so we've focused energy on this. Once we've fixed
the extra-symbols problems and are confident in it, this validation can be
upgraded to validate that too.
Half of rdar://problem/40431434.
Introduce a command-line option to visualize the complete set of output
request dependencies evaluated by a particular compile action. This is
exposing existing visualization facilities to the (-frontend) command line.
The signature here changed, but the old arguments implicitly converted
to the parameter types for the wrong overload, so we didn't notice.
Add a simple test that would have at least caught this for the driver.
It was only used in a few tests. Those tests now use -emit-sil instead
of -emit-silgen, with some functions marked @_transparent and a few
CHECK: lines changed now that the mandatory optimizations get to run.
This may help us reproduce a failing build when all we have is a build
log, and will become much more important in batch mode when we
/really/ need to know what ended up in what batch.
For now, this doesn't include /output/ filelists, because David's
about to mess with that code anyway to make things better around
supplementary outputs in batch mode. There is one weirdness there,
though, which is that ArgsToFrontendInputsConverter peeks at the
outputs to see whether we're doing single-threaded or multi-threaded
WMO.
This means moving the output path into SupplementaryOutputPaths, and
using the same sort of diagnostic dispatching that serialized
diagnostics use. This is part of what's needed to run the migrator
in batch mode.
Tiny start-up time optimization noticed while looking at how we do
PrettyStackTraceProgram. Also add PrettyStackTraceProgram to a few
more of our testing tools, via the new PROGRAM_START macro.
Before this patch, we have one flag (KeepSyntaxInfo) to turn on two syntax
functionalities of parser: (1) collecting parsed tokens for coloring and
(2) building syntax trees. Since sourcekitd is the only consumer of either of these
functionalities, sourcekitd by default always enables such flag.
However, empirical results show (2) is both heavier and less-frequently
needed than (1). Therefore, separating the flag to two flags makes more
sense, where CollectParsedToken controls (1) and BuildSyntaxTree
controls (2).
CollectingParsedToken is always enabled by sourcekitd because
formatting and syntax-coloring need it; however BuildSyntaxTree should
be explicitly switched on by sourcekitd clients.
resolves: rdar://problem/37483076
